Since logistic regression is a linear model the same measures are used as usual. A Correlation coefficient is poor because 2 IVs at a time are considered not all together as VIF AND tolerance do. I don't know what a multinomial IV could be but I suspect that your computation was.incorrect. follow the usual linear model procedures and you should be ok.. Best wishes, David Booth

In addition to David E Booth's comments: I think a VIF-score of 10 is a too high threshold value for detecting multicollinearity (and I'm not alone in this assessment). Problems start already at 5 or lower in my experience :-)

IMO, people often become far too concerned about multicollinearity. Here are a couple of blog posts that explain why I say that. (The first one is rather satirical.)

• https://davegiles.blogspot.com/2011/09/micronumerosity.html
• https://statisticalhorizons.com/multicollinearity

For more interesting commentary on this issue, try a Google search on .

Having said that, do agree with David Eugene Booth that if you are going to look at measures of multicollinearity, you want to look at VIF and tolerance (and SE inflation, etc.), not at pair-wise correlations (or other measures of association). See my 12-Mar-2021 post in this thread for an example where the pair-wise correlations are all moderate at best despite the presence of complete linear dependence. (Search for the name Dallal to find my post.)

https://www.researchgate.net/post/What_can_cause_high_multicollinearity_when_predictors_correlate_only_slightly

HTH.

Bruce Weaver Thank you for the resources... my main confusion stems from the fact my independent variables are categorical (for example, one of those variables is income and it has 4 different levels that I dummy coded so I can run the linear regression) which isn't as straightforward as if they were continuous ...

Christer Thrane thank you for your insight... would the cut-off score at 5 still be applicable even when dealing with dummied variables, I'm assuming that since dummy variables stem from one initial categorical variable, there's bound to be some level of multicollinearity just for that and hence maybe the higher cut off of 10 I have seen on websites like https://statistics.laerd.com/. (my assumption regarding the impact of the variables' nature on the VIF is based on my own mental gymnastics so I don't really have any resources to back it up :)

The VIF and tolerance statistics you get from the REGRESSION command in SPSS are valid only for quantitative or dichotomous variables. For categorical variables with 3+ categories, you'll need to look at the generalized VIF (gVIF). You can read about it in these notes by John Fox:

https://socialsciences.mcmaster.ca/jfox/Courses/soc740/lecture-7-notes.pdf

AFAIK, SPSS still has not incorporated the gVIF, so you may have to use R.

HTH.

Bruce Weaver yep hence why I dummy coded my categorical variables with 3+ categories to use SPSS (I have very limited knowledge in R) as I understood the dummy coding would help fix that problem and allow me to use SPSS... I am now playing around in R and trying to find the code to use .. I can't find a straightforward GVIF code but I read that it's similar to VIF and R will just compute the Generalized VIF if it sees I have categorical variables (?)

Hello Meryem Talbo. Converting a single factor variable into a set of k-1 dummy variables does not solve the problem. See slide 9 in John Fox's slides.

• https://socialsciences.mcmaster.ca/jfox/Courses/soc740/lecture-7-notes.pdf

He addresses this very situation in the first bullet point.

• Variance-inflation factors are not fully applicable to models that include related sets of regressors, such as dummy regressors constructed from a polytomous categorical variable or polynomial regressors

well, I ended up finding a way to use R and GVIF (pretty much just ran a regression then used VIF from the car package) and all my values are between 1.02 and 1.40)... guess no multicollinearity here.... Thank you all for your help!

this video was pretty helpful as well: https://www.youtube.com/watch?v=9Vz-CY3Gtmw